Transcript PPT - Big Data Open Source Software and Projects

Big Data Open Source Software
and Projects
ABDS in Summary XIII: Level 14A
I590 Data Science Curriculum
August 15 2014
Geoffrey Fox
[email protected]
http://www.infomall.org
School of Informatics and Computing
Digital Science Center
Indiana University Bloomington
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
HPC-ABDS Layers
Message Protocols
Distributed Coordination:
Security & Privacy:
Monitoring:
IaaS Management from HPC to hypervisors:
DevOps:
Interoperability:
Here are 17 functionalities. Technologies are
File systems:
presented in this order
Cluster Resource Management:
4 Cross cutting at top
Data Transport:
13 in order of layered diagram starting at
SQL / NoSQL / File management:
bottom
In-memory databases&caches / Object-relational mapping / Extraction Tools
Inter process communication Collectives, point-to-point, publish-subscribe
Basic Programming model and runtime, SPMD, Streaming, MapReduce, MPI:
High level Programming:
Application and Analytics:
Workflow-Orchestration:
Hadoop (MapReduce)
• Key algorithm supported by Hadoop for processing of large
data sets in a parallel, distributed fashion
• A two step process: data filtering and sorting (map) and data
summarization (reduce)
• Hadoop has become ubiquitous, used by more than half of
Fortune 50 companies. Installations are enormous, Facebook
stated in 2012 that they added 0.5 PB daily to their cluster.
• MapReduce concept originated at Google (2004), Hadoop
implementation grew out of Nutch search engine, largely
developed at Yahoo
• http://www.revelytix.com/?q=content/hadoop-ecosystem
describes many components of Hadoop ecosystem
Apache Spark
• Apache Spark is a general purpose computing framework
aimed at large data analytics applications.
• Spark APIs allow for programming in Java, Python or Scala,
and are meant to be simple to understand and use to support
rapid development.
• Spark includes a set of tools to integrate tasks such as
streaming, querying, graphing and machine learning.
• Spark is a young project, only becoming an Apache top level
project in February of 2014. Execution speed is a primary goal,
with some types of applications running up to 100 times
faster with Spark than with Hadoop.
Twister
• MapReduce programming model has simplified the implementations of many data
parallel applications. The simplicity of the programming model and the quality of
services provided by many implementations of MapReduce attract a lot of
enthusiasm among parallel computing communities.
• From the years of experience in applying MapReduce programming model to
various scientific applications we identified a set of extensions to the programming
model and improvements to its architecture that will expand the applicability of
MapReduce to more classes of applications. Twister is a lightweight MapReduce
runtime developed at Indiana University by incorporating these enhancements.
•
•
•
•
•
•
•
•
•
•
•
BSD style license http://www.iterativemapreduce.org/
Distinction between static and variable data
Configurable long running (cacheable) map/reduce tasks
Pub/sub messaging based communication/data transfers
using ActiveMQ
Efficient support for Iterative MapReduce computations
(much faster than Hadoop or Dryad/DryadLINQ)
Combine phase to collect all reduce outputs
Data access via local disks
Lightweight (~5600 lines of Java code)
Support for typical MapReduce computations
Tools to manage data
Automatically recover from faults when FaultTolerance is
enabled
Stratosphere / Apache Flink
• Originally developed in Germany but now Apache Flink
http://flink.incubator.apache.org/. The Apache Flink stack consists
of
– Programming APIs for different languages (Java, Scala) and paradigms
(record-oriented, graph-oriented).
– A program optimizer that decides how to execute the program for good
performance. It decides among other things about data movement and
caching strategies.
– A distributed runtime that executes programs in parallel distributed
over many machines.
Flink runs independently from
Hadoop, but integrates
seamlessly with YARN (Hadoop's
next-generation scheduler).
Various file systems (including
the Hadoop Distributed File
System) can act as data sources.
Microsoft Reef
• http://www.reef-project.org/
The Retainable Evaluator Execution Framework
• This is a set of tools that building on Apache Yarn or Mesos (or Google Omega, and
Facebook Corona) allow one to implement systems with capability of Hadoop and
Spark
• On these resource managers, REEF provides a centralized control plane abstraction
that can be used to build a decentralized data plane for supporting big data
systems, like those mentioned below.
– Special consideration is given to graph computation and machine learning applications, which
require data retention on allocated resources, as they execute multiple passes over the data.
i.e. Reef supports iteration
• More broadly, applications that run on YARN will have the need for a variety of
data-processing tasks e.g., data shuffle, group communication, aggregation,
checkpointing, and many more. Rather than re-implement these for each
application, REEF aims to provide them in a library form, so that they can be reused
by higher-level applications and tuned for a specific domain problem.
Apache Hama
• https://hama.apache.org/
• Apache Hama is a distributed computing framework based on Bulk
Synchronous Parallel computing (BSP) techniques for massive scientific
computations e.g., matrix, graph and network algorithms.
• BSP is used in most MPI programs and involves coordinated communication
separating computation (map) phases
• Hama was inspired by Google's Pregel large-scale graph computing
framework and is very similar in utility to the better known iterative
MapReduce systems but directly supports Graph applications
These types of plots are well known
for Iterative MapReduce
Hama much faster than Mahout on
parallel K-means
Apache Giraph
• http://giraph.apache.org/
• Apache Giraph is an iterative graph processing system built for high
scalability, to perform graph processing on big data.
• Scalability and speed is impressive, for example Facebook used
Giraph to analyze a trillion edges in 4 minutes.
• Giraph originated as the open-source counterpart to Pregel, the
graph processing architecture developed at Google and described in
a 2010 paper. Both systems are inspired by the Bulk Synchronous
Parallel model of distributed computation introduced by Leslie
Valiant.
• Giraph adds several features beyond the basic Pregel model,
including master computation, sharded aggregators, edge-oriented
input, out-of-core computation, and more.
Pegasus
• Apache License http://www.cs.cmu.edu/~Pegasus/
• http://www.cs.cmu.edu/~ukang/papers/PegasusICDM2009.pdf
• PEGASUS is a Peta-scale graph mining system, fully written in Java. It runs in
parallel, distributed manner on top of Hadoop.
• PEGASUS provide large scale algorithms for important graph mining tasks
such as: Degree, PageRank, Random Walk with Restart (RWR), Radius, and
Connected Components
• Graph Mining is an area of data mining to find patterns, rules, and
anomalies of graphs. Graphs or networks are everywhere, ranging from the
Internet Web graph, social networks (FaceBook, Twitter), biological
networks, and many more.
• Finding patterns, rules, and anomalies have numerous applications
including, but not limited to, the followings:
–
–
–
–
Ranking web pages by search engine
'viral' or 'word-of-mouth' marketing
Patterns of disease with potential impact for drug discovery
Computer network security: email/IP traffic and anomaly detection